Recognize and align lyrics

The Lyrics page writes timed lyrics onto a track named Lyrics, the track Prest Server’s REAPER plugin shows lyrics from. It has two tools, each a component of its own:

• Lyrics recognition writes down what is sung, as a draft for you to correct. About 30 MB, plus the Whisper model the first time you use it.
• Lyrics alignment times your words to the singing and writes them to the track. About 80 MB.

If you already have the words, you only need alignment. Click Install beside a tool on the page, or see Add the tools you need.

Use the vocal stem

Both tools need the voice on its own. Separate the song first and select its vocals: with the band still in, recognition hears words that are not there, and alignment misplaces whole lines. Use the full vocals, backing included; a lead-only (karaoke) stem aligns worse.

Recognize the words

1. Select the vocal stem’s item.
2. Open the Lyrics page.
3. Choose the language sung, or leave Detect language.
4. Choose the Whisper model. large-v3, the default, is the most accurate, and downloads about 3 GB the first time. large-v3-turbo is faster and nearly as good. tiny, base, small, and medium are smaller, faster, and less accurate.
5. Click Recognize.

Expected result: the words appear in the page’s editor, one line per sung phrase. Nothing in the project changes yet.

If the editor already holds words you typed, they stay, and the draft waits behind Use recognized words. Clicking it asks Replace your words? first.

If the item has no singing, the page says No singing found in this item: select its vocal stem.

Correct and align

1. Correct the words in the editor, or type or paste your own. Lines stay as you break them.
   • Labels in square brackets, such as [Chorus], are left out.
   • Words in parentheses, such as (ooh yeah), are backing vocals: they are not timed, and go onto the track directly after the word before them.
2. Click Align lyrics.

Expected result, in one undo step: the Lyrics track holds a MIDI item named “Aligned lyrics”, with one lyric event per word, at the time it is sung. A word that starts a line begins with /, the karaoke file convention, so the lines come back as you wrote them. REAPER’s lyrics view shows them, and they travel with the project. PresTools makes the track at the bottom of the track list if it isn’t there.

Aligning again replaces the lyrics items that start under the selected item. Lyrics for other songs on the same track stay.

The editor shows the words already on the Lyrics track for the selected item. Until you align, what you type stays in the editor only, and the page says not on the track yet.

Language

One setting for both tools, remembered. With Detect language, recognition guesses once per song from the singing, and alignment reads the language from your words. Guessing from the singing can confuse close languages, such as Ukrainian and Russian: choose the language when you know it.

Alignment works in English and over 60 other languages. Recognition offers every language Whisper knows.

Import an UltraStar file

An UltraStar .txt file already has its words timed and pitched, so it goes straight onto the Lyrics track. No selection and no component are needed.

1. On the Lyrics page, click Import UltraStar .txt: notes and words, the button in the page’s header beside How this works.
2. Choose the file.

Expected result, in one undo step: a MIDI item on the Lyrics track, starting at the file’s #GAP from the start of the project, named after the file. Each note is at its pitch, with its syllable as a lyric event where it starts, and the file’s line breaks are kept. Each kind of note has its own MIDI channel: normal 1, golden 2, freestyle 3, rap 4, golden rap 5. For a duet, the second singer goes on a track named Lyrics 2.

PresTools reads files in UTF-8, and older ones in the encoding their #ENCODING line names, or Windows-1252 without one. A file without #BPM can’t be placed in time, and the page says so. Move the item afterward if the song starts later in the project.

What it runs on

Both tools use the GPU unless you turn it off in Settings. Whisper runs with MLX on Apple silicon, and with CUDA or ROCm on NVIDIA and AMD GPUs. Apart from Whisper on Apple silicon, a run the GPU can’t do finishes on the CPU. Hover over Recognize or Align lyrics to see where each will run.

Next: Script the tools.